A Non-blocking Commitment Protocol

Abstract

A commitment protocol is one that ensures that at least some sites of a multi-site transaction do not block in spite of any single failure. This paper describes a quorum-based non-blocking commitment protocol that also subsumes the functions of termination and recovery protocols. The protocol survives any single site crash or network partition provided that the failure is not falsely detected. The protocol is correct despite the occurrence of any number of failures. and whether or not failures are falsely detected. When there is no failure. the protocol requires three phases of message exchange between the coordinator and the subordinates and requires each site to force two log records. Read-only transactions are optimized so that a read-only subordinate typically writes no log records and exchanges only one round of messages with the coordinator. Sites can forget the transaction after it terminates everywhere. Finally. a fundamental result about quorum-based commit protocols is uncovered: they are effective only for transactions involving more than three sites. Copyright © 1989 Dan Duchamp This work was supported by IBM and the Defense Advanced Research Projects Agency. ARPA Order No. 4976. monitored by the Air Force Avionics Laboratory under Contract F33615-84K-1520. The views and conclusions contained in this document are those of the author and should not be interpreted as representing the ofTicial poliCies. either eA-pressed or implied. of any of the sponsoring agencies or of the United States Government. lAuthor's current address: Computer Science Department. Columbia University. New York. NY 10027. The former Computer Science Department at Carnegie-Mellon University is now called the School of Computer Science.